Algorithms and Data Structures Lecture XIV

1. Algorithms and Data StructuresLecture XIV Simonas Šaltenis Nykredit Center for Database Research Aalborg University simas@cs.auc.dk

2. This Lecture • Introduction to computational geometry • Algorithms • Points and lines in 2D space • Line sweep technique • Closest pair of points using divide-and-conquer • A peek at data structures for mutidimensional range searching

3. Computational Geometry • The term first appeared in the 70’s • Originally referred to computational aspects of solid/geometric modeling • Later as the field of algorithm design and analysis of discrete geometry • Algorithmic bases for many scientific & engineering disciplines • GIS, robotics, computer graphics, computer vision, CAD/CAM, VLSI design, etc.

4. Geometric Objectsin Plane • Point: defined by a pair of coordinates (x,y) • Segment: portion of a straight line between two points – their convex combinaiton • Polygon: a circular sequence of points (vertices) and segments (edges) between them B A

5. Some Geomtric Problems • Segment intersection: Given two segments, do they intersect? • Simple closed path: Given a set of points, find a non-intersecting polygon with vertices on the points

6. Some Geomtric Problems (2) • Inclusion in polygon:Is a point inside or outside a polygon?

7. Segment Intersection • Test whether segments (a,b) and (c,d) intersect.How do we do it? • We could start by writing down the equations of thelines through the segments, then test whether thelines intersect, then ... • An alternative (and simpler) approach is based in thenotion of orientation of an ordered triplet of pointsin the plane

8. Orientation in the Plane • The orientation of an ordered triplet of points in the plane can be • counterclockwise (left turn) • clockwise (right turn) • collinear (no turn) counterclockwise (left turn) collinear (no turn) clockwise (right turn)

9. Intersection and Orientation • Two segments (p1,q1) and (p2,q2) intersect if and only ifone of the following two conditions is verified • General case: • (p1,q1,p2) and (p1,q1,q2) have differentorientations and • (p2,q2,p1) and (p2,q2,q1) have differentorientations • Special case • (p1,q1,p2), (p1,q1,q2), (p2,q2,p1), and(p2,q2,q1) are all collinear and • the x-projections of (p1,q1) and (p2,q2) intersect • the y-projections of (p1,q1) and (p2,q2) intersect

10. Orientation Examples • General case: • (p1,q1,p2) and (p1,q1,q2) have differentorientations and • (p2,q2,p1) and (p2,q2,q1) have differentorientations

11. Orientation Examples (2)

12. Orientation Examples (3) • Special case • (p1,q1,p2), (p1,q1,q2), (p2,q2,p1), and (p2,q2,q1) areall collinear and • the x-projections of (p1,q1) and (p2,q2) intersect • the y-projections of (p1,q1) and (p2,q2) intersect

13. Computing the Orientation • slope of segment (p1 ,p2 ): s = (y2-y1)/(x2-x1) • slope of segment (p2 ,p3 ): t = (y3-y2)/(x3-x2) • Orientation test • counterclockwise (left turn): s < t • clockwise (right turn): s > t • collinear (no turn): s = t

14. Computing the Orientation (2) • The orientation depends on whether the following expression is positive, negative, or null(y2-y1)(x3-x2) - (y3-y2)(x2-x1) = ? • This is a cross product of two vectors

15. Determining Intersections • Given a set of n segments, we want to determine whether any two line segments intersect • A brute force approach would take O(n2) time (testing each with every other segment for intersection)

16. Determining Intersections (2) • It can be done faster by using a powerfull comp. geometry technique, called plane-sweeping. Two sets of data are maintained: • sweep-line status: the set of segments intersecting the sweep line l • event-point schedule: where updates to l are required event point l : sweep line

17. Plane Sweeping Algorithm • Each segment end point is an event point • At an event point, update the status of the sweep line & perform intersection tests • left end point: a new segment is added to the status of l and it’s tested against the rest • right end point: it’s deleted from the status of l • Only testing pairs of segments for which there is a horizontal line that intersects both segments • This might be not good enough. It may still be inefficient, O(n2) for some cases

18. Plane Sweep Algorithm (2) • To include the idea of being close in the horizontal direction, only test segments that are adjacent in the vertical direction • Only test a segment with the ones above and below (predecessor and successor, rings a bell…) • the ”above-below” relationship among segments does not change unless there is an intersection • New “status”: ordered sequence of segments

19. Plane Sweeping Algorithm (3) b d c e a a ba bcb bcad bcd cd ecd ed e

20. Pseudo Code AnySegmentsIntersect(S) 01T ® 02 sort the end points of the segments in S from left to right, breaking ties by putting points with lower y-coords first 03 for each point p in the sorted list of end points do 04 if p is the left end point of a segment s then 05 Insert(T,s) 06 if (Above(T,s) exists and intersects s) or (Below(T,s) exists and intersects s) then 07 return TRUE 08 if p is the right end point of a segment s then 09 if both Above(T,s) and Below(T,s) exist and Above(T,s) intersects Below(T,s) then 10 return TRUE 11 Delete(T,s) 12 return FALSE

21. Implementing Event Queue • Define an order on event points • Store the event points in a priority queue • both insertion or deletion takes O(logn) time, where n is the number of events, and fetching minimum – O(1) • For our algorithm sorted array is enough, because the set of events does not change. • Maintain the status of l using a binary search treeT • the up-to-down order of segments on the line l <=> the left-to-right order of leaves in T • segments in internal nodes guide search • each update and search takesO(log n)

22. Status and Structure l Sm Sl Sk Si Sj Sk Sl Si T Sj Sm Sl Si Sk Sj

23. Running Time • Sorting the segments takes O(n log n) time • The loop is executed once for every end point (2n) taking each time O(log n) time (e.g., red-black tree operation) • The total running time is O(n log n)

24. Closest Pair • Given a set P of N points, findp,q Î P, such that the distanced(p, q) is minimum • Algorithms for determining theclosest pair: • Brute Force O(n2) • Divide and Conquer O(n log n) • Sweep-Line O(n log n)

25. Brute Force • Compute all the distances d(p,q) and select the minimum distance • Running time O(n2)

26. Divide and Conquer • Sort points on the x-coordinate and divide them in half • Closest pair is either in one of the halves or has a member in each half

27. Divide and Conquer (2) • Phase 1: Sort the points bytheir x-coordinate • p1 p2 ... pn/2... pn/2+1 ... pn

28. Divide and Conquer (3) • Phase 2:Recursively compute closestpairs and minimum distances, • dl, drin • Pl = {p1, p2, ... , pn/2 } • Pr = {pn/2+1, ..., pn } • Find the closest pair andclosest distance in centralstrip of width 2d, where • d = min(dl, dr) • in other words...

29. Divide and Conquer (4) • Find the closest ( ,) pairin a strip of width 2d,knowing that no ( , ) or( , ) pair is closer than d

30. Combining Two Solutions • For each point p in the strip,check distances d(p, q),where p and q are of differentcolors and: • There are no more than four such points!

31. Running Time • Sorting by the y-coordinate at each conquering step yields the following recurrence

32. Improved Algorithm • The idea: Sort all the points by y-coordinate once • Before recursive calls, partition the sorted list into two sorted sublists for the left and right halves

33. Running Time • Phase 1: • Sort by x and y coordinate: • O(n log n) • Phase 2: • Partition: O(n) • Recur: 2T(n/2) • Combine: O(n) • T(n) = 2 T(n/2) + n = O(n log n) • Total Time: O(n log n)

34. Multidimensional DS • Multidimensional Data Structures are used to answer queries on the set of multidimensional points • Range query ([x1, x2], [y1,y2]): find all points (x, y) such that x1<x<x2 and y1<y<y2

35. 1D Range Searching • Find x, such that x1<x<x2 • Can be done with balanced binary search trees (e.g., red-black trees) in O( logn + k) time, where k is the size of the answer T k elements x1 x2

36. Range Trees • Build a binary tree T on x-coordinates • With each node v, associate a binary tree Tv that stores all the points in the subtree of T rooted at v. Organize Tv on y-coordinates. • Space: O(n logn) • Range query: O( log2n + k)

37. Quadtrees • Quadtrees – partition tree • Linear space • Good average query performance b 1 3 2 a 3 2 4 4 c d e 4 1 3 3 a e b d f

38. Next Lecture • Introduction to NP-Complete